Item

Abstract

Global and regional climate models' prediction accuracy is limited, often with systematic biases between the model output and observed conditions. The present research aims to minimize the biases between the maximum temperature simulated by a global (HadGEM3) and a regional (WRF) climate model for the period of 2006–2014, based on the ERA5 reanalysis data, used as a proxy for observations (calibration period 1981–2000). For the bias correction, a new approach with the TIN-Copula method (combination of Triangular Irregular Networks-TINs and Copulas), suitable for gridded datasets is used. The methodology applies TINs to define the grids to which the corrections are made, and copulas for analysing the dependence between the reanalysis and simulated data during the calibration period. The dependence structure is then used for the bias adjustment. Our methodology is applied to the Middle East and North Africa MENA-CORDEX domain, where extreme heat conditions prevail, which are projected to accelerate during the 21st century. Besides to the entire area, a detailed analysis was performed for six subregions. Our results indicate that the maximum temperature output of both models diverges from the ERA5 reanalysis, especially during summer. In several subregions and seasons, discrepancies are highest for the HadGEM3 model, which has a much coarser resolution than WRF. The effectiveness of the TIN-Copula method was tested spatially, temporally and based on two relevant climate indices [daily maximum temperature (TXx) and warm spell duration index (WSDI]), showing that it minimizes model biases.